Rails Agent Skills
A curated library of 41 executable skills and 9 workflows that teach AI coding tools how to plan, test, implement, document, and review Rails work using production-minded conventions.
Why the old way fails
Most AI coding assistants produce plausible-looking code that lacks tests, skips review, and ignores your team's conventions. The result: inconsistent quality, missed edge cases, and code that looks right but breaks in production.
The new mental model
Each skill is a SKILL.md file containing task-specific instructions, conventions, and hard gates that an AI agent reads and follows. They are not tutorials to read once. They are constraints that run every time.
One skill per concern. plan-tests chooses the first failing spec. write-tests writes it. code-review reviews it. No monolithic guides.
Hard gates enforce discipline. The Tests Gate means no implementation code can exist until the test exists, runs, and fails for the right reason.
Each skill names the next skill to load in its Integration table. write-tests → implement → write-yard-docs → code-review. Skills are building blocks; workflows are the value.
How it works
The most-used daily workflow. Every feature, bug fix, and refactor follows the same repeatable cycle. The agent cannot skip a checkpoint without your explicit approval.
load-context
Sync schema, routes, patterns
plan-tests + write-tests
Choose and write the first failing spec
Checkpoint: Test Feedback
You approve the test before any code is written
Checkpoint: Implementation Proposal
You approve the approach before implementation
Implement → Test passes → Refactor
Minimal code, green test, clean up
Docs → Review → PR
YARD, self-review, merge
The library
Each skill has a SKILL.md entry point. Some have companion files for examples, test templates, patterns, or heuristics. Workflows chain skills into full development loops.
Concrete example
See the difference between a naive request and a skill-guided request. The skill enforces a multi-step process that protects your production database.
"Add a status column to orders with default 'pending' and not null."
AI generates:
"Add a status column to orders. Follow review-migration."
AI generates:
Integration
Use MCP for autonomous tool calls, chat commands for explicit control, or the CLI for manual installation and evaluation.
The recommended path. The agent discovers skills through list_skills and loads them on demand via use_skill. Context stays small.
Force the agent to use a specific skill. In Cursor or Windsurf, prepend with @. In Claude Code, describe the goal and the agent loads the right skill.
Install specific skills into your agent host. Pin to release tags for reproducible installs. Works with Claude, Cursor, Windsurf, and Codex.
What to do next
The fastest way to experience the difference: pick a small feature, run load-context, then plan-tests. Watch the agent stop at checkpoints, confirm boundaries with you, and only then write code.